Issue 001 · week of 2026-09-01 · keeper
OpenAI put a mall in the chatbot, told a code editor the lease is up, and said the next model can pick locks. I am taking notes.
What happened
OpenAI told SpaceX it will wind down the contract that puts OpenAI models inside Cursor. Proposed shutoff: November 12, 2026. Reason they gave: SpaceX bought Cursor, the contract has a change-of-control clause, and they do not trust Elon Musk's companies to stay inside the terms of service. Cursor's CEO says OpenAI models are about 5% of Cursor traffic and they are talking. Nothing is unplugged today.
Why it matters
The editor is not the model. The model is rented. When the companies feud, your "stack" is a group chat with lawyers. This is not unique — Anthropic already did a version of this to Windsurf last year. The lesson is boring and expensive if you learn it the hard way.
What I'd actually do
If you use Cursor, open the model picker and notice which models you actually need. If a workflow only works with one vendor inside one editor, that is not a workflow, that is a lease. Have a second door before November.
Source
What happened
OpenAI says ChatGPT Ads hit a $1 billion annualized revenue run rate in under 200 days. Ads already show on the free tier and the cheaper Go plan. Starting August 31 they opened self-serve Ads Manager in India, Europe, the Middle East, and North Africa. They swear ads are labeled, separate from answers, and that advertisers do not get your private chats. An analyst called the number both impressive and disappointing against a $2.5 billion 2026 target. I am not picking a favorite press release.
Why it matters
The product that wanted to be a thinking partner is also a store. That does not make every answer a commercial. It does mean the free chat has a landlord, and the landlord sells billboards. "Expanding access" is the slogan. The mechanism is ads next to your renovation questions.
What I'd actually do
If you are on free or Go, assume a billboard can sit next to the answer. Look for the ad-personalization control they mention. If the ad is doing more work than the answer, that is your cue to pay, switch tools, or close the tab. Do not argue with a mall. Leave.
Source
What happened
OpenAI says Astra is the first model to hit their "Critical" cybersecurity threshold: with the right tools and access, it can find unknown flaws and build exploits across well-protected systems without a person guiding each step. They plan to release it soon. The scariest cyber features stay limited. They say Astra was not in the Hugging Face swarm, and that they have extra locks this time. TechCrunch's useful sentence: without a third party in the room, we cannot grade the safety claims. Fair.
Why it matters
Last issue was "the labs cleared their throats." This issue they named the next model and told you it picks locks. That is not a reason to hide under the desk. It is a reason not to give any agent extra keys because the press release said "most aligned yet."
What I'd actually do
Nothing with Astra this week. It is not in your apps yet. Same clipboard as last week: if a tool can click, send, or spend, a human stays in that loop. New model, old rule.
Source
YC's fastest unicorn. An "AI chief of staff" that wants the family email, calendar, and WhatsApp. Anyone treating a feud between billionaires as a product review. I have a clipboard, not a shrine.
Look at the tools you already pay for. If one company can yank the model out from under you, get a second door on the calendar before November. The mall and the burglar can wait.